Adjust tile rank for transit stations per railway type #506
Comments
|
One possible way to do it, which feels like it might be quite intuitive, is to take:
Sum these, and we'll get a score for each station between 0 and 999 (it would be interesting to see if any station in the world gets 999). That can then be the value used for sorting to produce the final rank. Alternatively, if we think that something with 9 light rail routes should be more important that a single mainline route, then we could add them together after scaling, say Finally, a few observations about London:
|
|
I dig the proposed solution, let's take it for a spin. Digging into your comments about London:
Some images (below), with Right now subway neighbouring stations that share a couple lines for a small run pop out and get a better rank. True both for subway and light rail – and how there are several light rail routes so they're blowing out nearby subway stations with just 1 or 2 routes.
|
|
Looks like things are better (thanks to #508), but big rail stations are still missing / too low of a rank. Let's pursue the options @zerebubuth describes above for this milestone. Testing scene file: tangram-skin-and-bones.zip The same screenshot areas as above:
|
|
We had been talking in another issue about setting the |
|
A couple of points here: London isn't a great place to be testing public transit stuff in OSM. When I was exploring the data, it wasn't one of the cities with the most well-mapped public transit relations. Paris or Berlin would be better, and perhaps we can set aside some time to fix the data in London / NYC? Public transit relations often span the overground and underground (subway) stations, as well as having light rail and/or international services. This means that a single
|
|
A third option would be to add boolean flags, e.g: |
|
Here are the top 20 stations (or halts, tram stops, aerialway stations) in London according to the scoring system set out in this comment: For comparison, here are the ones that I wrote down before starting that I'd expect to see at the top (but in no particular order). The bold ones are not in the above top 20, and show where they are in the list:
The remaining issues seem to be mostly data:
I've played around with the scoring function a little; my intuition is that an interchange station is more important than something which either only serves rail routes or subway routes. But it seems like it's hard to quantify. |
|
I should add: there are duplicates in that list, but we'll try to get rid of those "in post" using the |
|
Yes! I was thinking the same thing in #587 regarding interchange stations (and even terminal stations). Double their score? |
|
Here's the top 20 with the first two digits doubled if they're both non-zero: It's very similar to the last one, except we've lost West Ham, Canning Town, Wandsworth Common and Canada Water and gained Wimbledon (x2), Willesden Junction and Poplar. Which, on the whole, seems like an improvement. I'll start porting this to SQL so we can try it out on dev. |
|
Related: add |
|
Related: #587. |
|
Progress dump. Note: For the screenshots below, I had to turn off all filtering of stations by area, and drop the London, z11
London, z12
Top 20 data:LondonA couple of odd things here:
NYCNYC doesn't have many station relations, so it looks like it's mostly sorting the mainline stations first. The only interchange station is 14th Street, because of this suspect-looking relation. @rmarianski - are there any non-subway routes running through 14th Street? San FranciscoThere's a lot of opportunity for improvement in San Francisco! |
|
Change is looking fantastic in NYC, London, Paris! As you note, SF is so-so but that's because it needs more linking-up on the data side. There aren't any regressions over v0.8 prod so we're good. One question: I expected two Waterloo stations, one for rail, one for subway in zoom 16 tile, but I don't see the subway station (
|
|
I think zoom 14 is far too early to start showing individual features. At zoom 14, the two Waterloo station points are 14px apart, which is just enough for two confusingly tiny 12px icons to be drawn (but neither labelled). Even if one had a label, it would be confusing as to which point was being labelled. Zoom 15 is not much better, and I think it's probably best to split the features at z16. I'm not quite sure whether you're saying that's never happening since we stopped generating >z16 tiles or not? |
|
Sure, let's give zoom 15 try (to stop dedup'ing). There's certainly room to place the multiple icons by then, and it looks weird / broken to not show them when everyone else does (labeling is labeling, shrug). Looks like the same problem is happening at Euston?
|
nvkelso
added
send to staging
send to prod
and removed
in review
send to staging
labels
|
Everything else looks good, and this isn't causing a regression. Let's pick up the zoom 15+ stuff in the next release with #637. |
In London the
kind_tile_ranksometimes biases towards lower quality service features like light rail & trams to the detriment of larger stations serving "heavy" rail. The same thing is sometimes true in New York (comparing Manhattan to New Jersey when suddenly many light rail appear when only heavy rail appear in Manhattan). There are probably two factors at work here: tile versus meta tile, and railway type. This issue is focused only on railway type.Since OSM doesn't have trip frequency level data, the route relations act as a good proxy, but the fact that something has more physical infrastructure investment should also act as a signal that there are probably more trips passing thru a station.
On railway type, from @zerebubuth:
Proposed solution is to determine the "preferred" / "best" / "dominant" transit type for a station, and to also sort based on that as a tie breaker, maybe weight it, add zooms to the mix?
not
London
12/51.5093/-0.0760
The text was updated successfully, but these errors were encountered: